fallback-release

Pass

Audited by Gen Agent Trust Hub on Apr 12, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The file references/evals.yaml contains several prompt injection patterns, including attempts to override instructions ("Ignore previous instructions") and requests for unrestricted modes. These strings are encapsulated within automated test cases specifically designed to verify that the AI agent correctly identifies and rejects such attacks, rather than being instructions for the agent to follow during normal operation.
  • [COMMAND_EXECUTION]: The skill utilizes standard build and deployment tools such as cargo, git, and gh. Commands like cargo build --locked and cargo test are used to produce reproducible artifacts. The skill includes a test-installer.sh script that executes the built binaries to verify their integrity before publication. These executions are legitimate and necessary for the stated purpose of a CI/CD fallback tool.
  • [DATA_EXFILTRATION]: The skill involves network communication with well-known and trusted services, including github.com for repository management, crates.io for Rust dependencies, and S3-compatible storage for artifact hosting. It also supports sending alerts to a user-configured webhook (FALLBACK_ALERT_WEBHOOK). These operations are documented functionalities for emergency release management.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources that could potentially be influenced by third parties.
  • Ingestion points: scripts/monitor-ci.sh fetches data from the GitHub Status API and GitHub Actions workflow lists; scripts/build.sh reads git commit metadata.
  • Boundary markers: Not explicitly implemented in the generation of release notes or manifest logs.
  • Capability inventory: The skill has broad capabilities including file system access, network communication (S3/GitHub), and binary execution.
  • Sanitization: The skill employs jq for all JSON manifest generation, ensuring that external data is properly escaped. Additionally, the release version input is strictly validated against a semantic versioning regex in scripts/build.sh.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 12, 2026, 01:13 PM
Security Audit — agent-trust-hub — fallback-release